S-052 — etf-sector-reversion-v1 (market-neutral residual reversion on 9 liquid sector ETFs)
ID: S052
Slug: etf-sector-reversion-v1
Card title: Market-neutral residual reversion on 9 sector ETFs
Card metrics: Horizon 21y | AnnRet -1.7%
Type: Stat-arb pillar — market-neutral, few-position / all-liquid exploration (sibling of S-038/D-008).
Date registered: 2026-07-15
Status: ONGOING exploration — failed; NOT frozen, NOT counted, NOT live (CLAUDE.md §2A).
Motivation: S-038 (stock residual reversion) is market-neutral and high-Sharpe but trades ~188 names, some
illiquid. Question: can the same edge be had with a handful of liquid ETFs?
Idea
Run the identical Avellaneda-Lee methodology (WIN=60, OU s-score, S_OPEN 1.25 / S_CLOSE 0.5, kappa < 30d, 10bps/side) on the 9 SPDR sector ETFs, each's residual taken vs SPY (single factor). Long oversold-residual sectors, short overbought, beta-hedged. ≤9 positions, all liquid. Same 2023+ holdout.
Finding (2026-07-15) — failed: the reversion edge does NOT survive at the sector level
Engine the exploration workspace (avg 3.7 open, turnover 17/yr, beta ~0):
| direction | in-sample ≤2022 | holdout 2023+ | net/yr out-of-sample (OOS) |
|---|---|---|---|
| Reversion (long oversold) | −0.06 | −0.75 | −6.5% |
| Momentum (long strong) | −0.37 | +0.38 | +3.3% |
- Reversion fails structurally — sectors don't mean-revert vs the market, they trend (the "residual duality" finding: momentum is sectoral). −0.75 OOS.
- Momentum is inconsistent — negative in-sample (−0.37), positive holdout (+0.38); the sign flips across regimes → not a reliable edge (and the OU bands are reversion-tuned, a poor momentum exit). The +0.38 OOS is the S-050 small-sample trap; not chased.
Read: the reversion edge genuinely lives in the breadth of stock-specific residuals; 9 sector ETFs can't replicate it, and the sector level is where momentum lives, not reversion.
Where next
Superseded by statarb-liquid-cap-v1 (un-numbered — its S-053 was released; dropped after the robustness
read) — keep the real stock-level reversion edge but restrict to the
most-liquid names (lower topn, since the universe is already dollar-volume-ranked) + a position cap. That
addresses the owner's two real constraints (fewer positions + liquid only) without abandoning the working edge.
ONGOING exploration; a failed read logged in the open. Do NOT count or freeze.